Experimental debt aversion and actual debt behavior

(1)

EXPERIMENTAL DEBT AVERSION AND ACTUAL DEBT BEHAVIOR

Abstract:

Potential students weigh the benefits and costs of entering higher education. They are confronted with an important consideration: the debt dilemma. On average, university graduates have high debt. If debt aversion, anomalous behavior regarding an unjust distaste for debt, is a factor in debt dilemma then there are potential opportunity costs, such as lower prospects for individuals and forgone human capital in society. For this thesis, an experiment examines the behavior of Dutch students to determine if they are debt averse. The results do not indicate that subjects are debt averse, however, a significant relationship is found when subjects are analyzed individually. Specifically, students who are debt averse have a lower fraction of debt in their monthly income (16.7%), in comparison to non-debt averse students.

Florian Korn (10459944)

Thesis Msc. Economics (15 ECTS)

Track: Behavioural Economics and Game Theory University of Amsterdam

Supervisor: Dr. Aljaž Ule

(2)

Statement of Originality

This document is written by Student Florian Korn who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

1. Introduction

When weighing the benefits and costs of a university education, the debt dilemma is an important consideration. While completing a bachelor and master program is expected to improve prospects, full-time education comes with explicit costs, such as tuition, living and housing expenses, as well as implicit costs, such as the opportunity to work and gain experience.

Dutch students accumulate an average of 528 euros debt each month (Van

der Werf, Schonewille, & Stoo, 2017). After graduation, the expected debt is 21.000 euro (“Beloftes”, 2018), which is a barrier for potential students. Some chose not to attend university, or to save money first, because they did not want to accumulate this level of debt (“Minder studenten”, 2016).

If potential students decide not to attend university because they overestimate the costs or they wrongfully underestimate the future benefits of education, then they are described as debt averse. If this anomalous behavior is an important factor in the debt dilemma, then they will weigh benefits and costs incorrectly resulting in the decision not to attend university. This comes at a cost for individuals and society, namely forgone knowledge as well as lost future tax

revenues (Cunningham, & Santiago, 2008).

The aim of this thesis is to test the internal and external validity of debt aversion in experiments. Debt aversion is defined as follows: when subjects need to smooth consumption, they are less willing to take on debt as they are to save. In other words, subjects deviate more from the optimal consumption smoothing path when they need to take on debt, in comparison to a treatment where they need to save to smooth consumption.

Internal and external validity are important to research. Internal validity affects whether the results are replicable. External validity means that results can be generalized to a population. Within experimental economics, there is a debate about the importance of both validities.

The importance of external validity and the possible trade-off with internal validity is a topic addressed by several significant papers. Schram (2005) describes the tension between internal and external validity and the artificial results created in the laboratory. He concludes that experimental economists focus more on internal validity than external validity, which allows researchers to test a specific hypothesis.

(6)

2

In contrast, Jiménez-Buedo and Miller (2010) argued that the internal and external validity of experiments are equally important and that there is no need to compromise between validities. Replication of experiments with slight variations should allow researchers to avoid potential problems that arise in observed results. Conclusions are stronger if similar results are found despite slight alterations.

In contrast, Levitt and List (2007) found that the behavior of subjects in experimental economics is altered by five factors: increased presence of moral consideration, the nature and extent of scrutiny of actions by others, the context of the decision, self-selection and the stakes of the experiment. A common assumption in economics is that individuals maximize utility.

However, this behavior can change in an experimental setting. Furthermore, Levitt and List (2007) argue that these experimental factors affect the level of altruism, fairness and conditional reciprocity thereby blurring possible external validity.

Meissner (2016) found debt aversion in an experiment with a borrowing and a saving treatment. In both treatments, current income is known, future income is uncertain and for each decision two outcomes are possible with equal probability. To compare treatments, total

expected income is equal. If subjects smooth their consumption, they receive the highest payoff. Meissner concludes that subjects significantly deviate from the optimum in both treatments. Additionally, subjects deviate more in the borrowing treatment than in the saving treatment. Therefore, it is concluded that subjects are debt averse.

To test the internal validity of the research of Meissner, whether debt aversion can be found among Dutch students, this thesis will use a simplified version of Meissner’s experiment. Furthermore, this paper will extend Meissner’s research by surveying the actual debt behavior of the subjects and testing the relationship between experimental debt behavior and actual debt behavior.

To test the external validity of experimental debt aversion the following research question is used:

To what extent does debt aversion affect real debt behavior among Dutch students?

The experiment in this thesis consists of four rounds (total duration of approximately 45 minutes). Each round consists of five consumption decisions. Subjects see their income stream

(7)

3

before each round starts and they are presented with either a decreasing income stream (saving treatment) or an increasing income stream (borrowing treatment). Income in a decision is

dependent on chance (a coin toss) and has two possible outcomes. Subjects make a consumption decision after their income for that decision is determined. Payout is maximized by smoothing consumption over the five decisions. The subjects need to form correct expectations of future income. Each treatment was repeated once, after round two a new treatment was introduced.

To analyze the experimental behavior, two optimality conditions are used: objective and subjective optimality. The objective optimality condition effectively smooths consumption according to expectations, whereas the subjective condition does so while taking the history (prior decisions) of a subject into account. Deviations from these conditions in one round give critical insight into the behavior of subjects. The median of these deviations indicates

overspending (positive median) or underspending (negative median). Debt averse behavior is found when subjects deviate more from one of the optimality conditions in the borrowing treatment (borrow too little) than in the saving treatment (save too little). Debt averse subjects are less willing to borrow in the borrowing treatment than they are to save in the saving treatment to smooth consumption. Overall, it is concluded that Dutch students are not debt averse, despite significant deviations from the subjective optimality condition.

To measure the effect of debt aversion on real behavior, two independent variables are introduced: “monthly debt increase” and “debt as a fraction of income”. It is concluded that debt averse students have a lower debt as a fraction of income (16,7% lower). Consequently, it is concluded that individual debt aversion affects real debt behavior.

The remainder of this thesis is organized as follows. In the next chapter, the related literature is discussed. Chapter 3 describes the methodology. Chapter 4 provides the results of the regression models. In chapter 5 the results are discussed and chapter 6 is the conclusion.

2. Related literature

This chapter will first describe consumption theory by assessing important models and describing the relationship of consumption with debt aversion. Secondly, the current status of intertemporal consumption smoothing experiments is examined. The third section defines several hypotheses.

(8)

4

2.1 Theory of consumption

To get a better understanding of consumption and consumption smoothing this section will discuss important theories. Furthermore, the effects of debt aversion on society will be discussed with a focus on students.

2.1.1 Introduction to consumption

Consumption is at the center of economics since the discipline’s origin (Smith, 1776). According to Keynes (1937), an important factor for the level of consumption is income. He used the

propensity to consume and the share of income used for consumption, to explain how a change in income will affect consumption. He argues that the average propensity to consume falls as income rises. Other determinants of current consumption are expectations about future income, lifestyle, attitude toward saving, a standard level of consumption, active saving strategies (e.g. pension savings), availability of consumer credit, new employment perspectives and fiscal conditions (e.g. anticipation effect of changing taxes) (Sulkunen, 1978).

Changes in income affect consumption (Heckman, 1974). Individuals are dependent on their parents (or other providers), before they are able to earn their own income. Furthermore, wages are low when individuals first enter the labor market. However, as careers evolve wages increase, affecting consumption and savings. When individuals retire they lose their labor income. Sustaining a reasonable level of consumption after retiring, is an important incentive to start saving. Parallel to consumption, there are multiple reasons to save such as becoming financially independent, preparing for significant expenses, such as a home or car, unforeseen expenses, emergencies and maintaining a good quality of life (Browning, & Lusardi, 1996). 2.1.2 Consumption models

There are many consumption models in economics such as Keynes’s general theory, relative-income hypothesis, life-cycle model, random-walk hypothesis and the permanent-relative-income hypothesis. Despite the historical relevance of these models, this section will focus on the life-cycle model and the permanent-income hypothesis because consumption smoothing is a key concept in these models.

2.1.2.1 Life-cycle hypothesis

Ando & Modigliani (1963) introduced the life-cycle hypothesis (LCH). This model explains the consumption patterns of individuals. The authors emphasize the functions of saving and

(9)

5

transferring purchasing power between life phases. Income is low (non-existent) before

individuals work. Labor income typically peaks in the last part of the working life. This income is lost when individuals retire. To smooth consumption over this life-cycle, consumers need to: borrow in the early stages of life, repay loans and save during their working life and spend the accumulated savings during retirement. To calculate the level of smooth consumption, the life-cycle approach uses a lifetime budget. This budget consists of future income and current wealth discounted to the current period. Because of the diminishing marginal utility of consumption, it is optimal to smooth consumption. Increasing consumption by one extra unit leads to a smaller increase in utility than the previous unit of consumption.

An important concept of this model is the dissaving (using savings for consumption) of the elderly to the point that they consume all of their accumulated wealth (Bernheim, 1987). But this is not reflected in reality for two reasons. Firstly, retired individuals accumulate

precautionary savings for unexpected expenses because they may surpass their life expectancy or they expect to have substantial medical expenses. Secondly, the elderly save more so that they can leave bequests to their children. Both explanations indicate that the LCH is not complete (Mankiw, 2012). Kimball (1990) argued that the amount of precautionary savings depends on the level of risk aversion and the resistance to intertemporal substitution. Higher risk aversion leads to higher precautionary savings, while greater resistance to intertemporal substitution results in lower precautionary saving.

2.1.2.2 Permanent-income hypothesis

Another consumption model is the permanent-income hypothesis (PIH) by Friedman (1957). Friedman did not focus on the life-cycle but on income fluctuations. He made a distinction between permanent income and transitory income. Permanent income is the normal level of income that households expect over their lives. Deviations from that level are the transitory income. Furthermore, Friedman distinguished between permanent and transitory consumption. Permanent consumption is steady and planned and transitory consumption is unexpected and irregular spending. He assumed that total consumption is independent of transitory income and that transitory consumption is independent of permanent income.

To maximize total lifetime utility, an individual must smooth consumption because of diminishing marginal utility (Similar to LCH). Transitory income fluctuates randomly affecting income in each period resulting in fluctuations in wealth if permanent consumption remains

(10)

6

stable. When transitory income is negative an individual must dissave to maintain a stable level of consumption.

There is mixed empirical evidence for the PIH. Hall and Mishkin (1982) found that consumption responds more to permanent changes than to transitory changes in income. Shapiro and Slemrod (2003) found supporting evidence for the PIH. They argue that individuals spread a temporary tax cut over a long time-period. On the other hand, Souleles (1999) found that people did not take into account a known income tax refund when making decisions about current consumption. Another important rejection of the PIH is the existence of possible liquidity constraints. If individuals are not able to lend money or take on credit then they can be restricted to a lower level of consumption when transitory income is negative (Mankiw, 2012).

2.1.3 Consumption smoothing

In addition to liquidity constraints, there are other factors that negatively affect the consumption smoothing of individuals.

In general, as the LCH and PIH predict, across the life-cycle there is a hump shaped pattern in income (Bernheim, 1987). To smooth consumption, individuals should take on debt in the early stages of life. Individuals depend on their future self to be able to repay this debt and so their future self is burdened with debt thus constraining their consumption with previous debt behavior.

Parker, Brunnermeier, & Papakonstantinou (2016) identify additional reasons for a lack of consumption smoothing. They investigated households that received a randomly-distributed stimulus and collected information about their daily purchasing behavior. They argue that consumption smoothing is negatively affected by a lack of financial planning. Parker et al. argue that financial planning is expensive. Another reason individuals spend too much directly after receiving their money is impatience. According to Parker et al., households that indicate that they

would rather enjoy today than to save for the future, smooth their income less effectively than

other households. Furthermore, they find that households with low liquidity have a high spending response, in comparison to high liquidity households. They emphasize that low liquidity in households is highly correlated with low income. Additionally, high earners can afford better financial advice than low earners. Parker et al. argue that the main determinant of consumption smoothing is low-income, which works its way through other variables. Parker et

(11)

7

al. do not find evidence that self-control or procrastination in households affects the smoothness of consumption.

Morduch (1995) emphasized that individuals and households are innovative in finding ways to protect themselves against risk, mainly unexpected changes in income, by smoothing consumption through credit, insurance, setting up long-term wage contracts and diversifying income to reduce the risk of transitory shocks.

2.1.4 Consumption smoothing and debt aversion

Hyperbolically discounting future utility provides a reason for the existence of debt. People pay a premium to extend costs to the future by paying interest on their debt. However, research

indicates a reluctance towards debt, known as debt aversion (Caetano, Palacios, & Patrinos, 2011). This anomaly negatively affects individuals’ ability to smooth their consumption in the early stages of life. If debt averse individuals are restricted to their current income, then prior saving is needed for large expenses, such as a house or car.

In particular, students, a portion of the population that expects higher earnings, is

confronted with a debt dilemma. If they are willing to complete their education fulltime, they can be submerged in high levels of debt in order to pay for basic needs, college tuition and housing. Cunningham & Santiago (2008) indicate that borrowing students are more likely to be enrolled fulltime, be unemployed, have high expectations about future income, live in dorms, and not receive financial support. Furthermore, demographic factors play a role. Among non-borrowers, black and Hispanic students were more likely to work full-time than white students indicating that white students receive more parental support. Additionally, non-borrowers are more likely to leave college without a degree. Cunningham & Santiago identify this as a lost opportunity for society. The authors make clear recommendations to address loan aversion and to reduce the need to borrow.

Walters, Erner, Fox, Scholten, Read, & Trepel (2016) support the existence of debt aversion. They state that it is anomalous behavior because it violates standard economic

intertemporal discounting. Walters et al. argue that debt aversion is common and appears to serve as a heuristic for financial behavior (debt is bad). Furthermore, they indicate that debt aversion is seen as a good trait in society.

Prelec and Loewenstein (1998) found debt aversion among consumers and workers. Consumers prefer to prepay for consumption. The negative aspect of paying directly is mitigated

(12)

8

by the idea of future free consumption benefits. This is contrary to the prediction that consumers prefer to pay for their marginal consumption. A similar argument can be made for workers who prefer to receive a salary after working.

On the contrary, Eckel, Johnson, Montmarquette, & Rojas (2007) did not find an effect of debt aversion on the demand for loans in postsecondary education (comparable to university). They focus on the role of a person’s attitude towards debt and their experience with debt. Eckel et al. indicate that subjects who are more experienced with debt are more willing to take on additional debt. Moreover, demographic factors of age, income, employment status, risk aversion, time preference (impatience) and ability show a significant relation with the willingness to use debt to finance education.

This research demonstrates that debt aversion is a common trait in humans. Students avoid high debt and consumers are willing to overcome direct costs to be able to consume freely in the future. It is clear that debt aversion negatively affects society if borrowing in the early stages of life is decreased.

2.2 Current experimental results of intertemporal models

This section will describe the current status of experimental economics regarding dynamic intertemporal life-cycle models.

Dynamic optimization problems such as intertemporal consumption and savings

problems have been researched extensively in experimental economics. Subjects tend to deviate from the optimal behavior. If subjects face a finite income stream they tend to overspend (save too little) in early decisions compared to the consumption smoothing path. It is a common belief that subjects learn through repetition or by observation. Subjects gravitate towards a more optimal consumption path, when they have opportunity to learn (Carbone, & Duffy, 2014).

Research in dynamic intertemporal consumption problems has been focused on saving behavior. Often subjects were confronted with a decreasing income stream (saving in the early decisions is required to smooth consumption), simulating the loss of income as a result of retirement. Within the majority of these experiments, subjects had a borrowing constraint. They could not spend more than their current income plus accumulated wealth.

Meissner (2016) allows for borrowing and uses two treatments that differ in income streams: an increasing income stream (borrowing treatment) and a decreasing income stream

(13)

9

(saving treatment). The experiment incentivizes consumption smoothing. Therefore, the optimal strategy is different for both treatments. In the increasing income stream, a subject must borrow in early periods and pay off debt when income is high. Facing the decreasing income stream an individual must save when income is high and dissave to maintain consumption. The total expected income is equal to allow for comparison of both treatments, which lead to the same expected level of optimal smoothing consumption. In contrast to previous claims, Meissner (2016) found that subjects underspend in the borrowing treatment. They do not borrow enough to smooth consumption. Furthermore, he finds that subjects overspend in the saving treatment, which is consistent with prior literature. Additionally, Meissner concludes that subjects are less

willing to borrow than they are willing to save in order to smooth consumption (2016, p.283). In

other words, subjects deviate more in the borrowing treatment than in the saving treatment. Therefore, he argues that subjects are debt averse.

Two explanations are given for debt aversion. Ballinger, Palumbo, & Wilcox (2011) indicate that the ability (bounded rationality or cognitive constraint) of a subject is the best predictor for optimal saving behavior. Furthermore, they find that subjects tend to improve with experience (this is not possible with perfect rationality). Brown, Chua, & Camerer (2009) identified a second explanation. Suboptimal behavior is caused by a hyperbolic time preference, in which subjects prefer consumption in the present. This would explain overspending in the early stages of the experiment. Furthermore, Thaler (1980) argues that it is hard to predict human behavior because people cannot be perfectly rational and they make systematical errors.

2.3 Hypotheses

This section describes the contribution of this thesis and three hypotheses.

The goal of this thesis is to test the external validity of experimental findings regarding debt aversion. As described in chapter one, there is an increasing interest in externally valid experiments. A simplified version of the experiment of Meissner (2016) will be used to replicate the results with a different pool of subjects: Dutch students. Furthermore, subjects are surveyed on multiple controlling factors (identified in the sections above) and their actual debt behavior.

(14)

10

1. Subjects overspend in early decisions of a round in comparison to the optimal consumption path in both treatments.

2. Subjects improve with experience and gravitate towards the optimal consumption path. 3. Subjects overall show debt averse behavior. They borrow (save) too little in the

borrowing (saving) treatment and deviate more in the borrowing treatment than in the saving treatment

Hypothesis 1 is tested to verify the results of previous literature (Carbone, & Duffy, 2014). Hypotheses 2 and 3 are tested to replicate the results of Meissner (2016). After these hypotheses are tested, previous literature is extended to test the external validity of this experiment.

3. Methodology

This chapter describes the methodology. First, the experiment is discussed in section 3.1. Secondly, in section 3.2 the experimental procedures are explained. Lastly, the experimental analysis is discussed in section 3.3.

3.1 Experiment

The experiment consists of four rounds. Each round has five decisions. Subjects choose their consumption with every decision. In total twenty decisions are made.

Subjects who participated in the experiment are Dutch students, who were chosen because they are entitled to a government loan due to their enrollment in a bachelor or master program. Participants receive payment based on their performance in the entire experiment. Their payment is equal to the total earnings in all rounds divided by two. Average total earnings equaled 495 euro cents, the highest earning subject collected 605 euro cents.

The final part of the experiment consists of a survey in which subjects are asked several control questions. These questions are identified as potentially important to actual debt behavior (Table 1). Furthermore, subjects were asked about their actual debt behavior. Importantly, 22 of 30 subjects have monthly increasing debt with an average increase of €298,50 each month. This debt is 27.5% of the monthly income on average. See Appendix 1 for the complete experiment, including the instructions.

(15)

11

n = 30 Number of males

(females)

Job (No job) Support from parents or government (No support)

18 (12) 28 (2) 19 (11)

Expect to be a high earner (Does not expect to be a high earner) University student (others HBO-student) The average monthly debt increase

The average percentage of income from debt

21 (9) 22 (8) €298,50 27,5%

Table 1: Survey summary 3.1.1 Income

Before a round starts the participants see their income stream. Table 2 shows income is in Experimental Currency Units (ECU). Income can take on two values in each decision. To determine the income for the current decision, a coin is tossed. Heads gives a lower income than tails. Before participants decide on their consumption they see their income. Therefore, current income (income of the current decision) is known. Future income is uncertain.

Borrowing treatment Saving treatment

Decision Heads Tails Heads Tails

1 4 ECU 6 ECU 0 ECU 2 ECU

Table 2: Treatments

3.1.2 Consumption decision

Participants are free to choose any level of consumption, number of ECU spent in each decision. They spend a positive amount of ECU to buy euro cents, see the conversion in Table 3.

Participants can spend part of their current income, saving the remaining income or repaying accumulated debt. Furthermore, subjects can spend more than their current income by using previous savings or increasing their debt.

(16)

12

Table 3: ECU euro cent conversion Table 4: Treasury balance 3.1.3 Treasury

To keep track of borrowing and saving a treasury is used. If total consumption is less than income, then the treasury is positive. The treasury is negative when consumption exceeds income. Therefore, if subjects spend more (less) ECU than their current income, their treasury decreases (increases). If they spend their total current income, their treasury is unaffected.

At the end of every round, when all five decisions are made, the treasury must balance (return to zero) to avoid unlimited borrowing (Meissner, 2016). If, after the final decision in every round, the treasury is positive (ECU not spent), it will balance by disregarding any amount. If a participant spends exactly their total income there is no need for treasury balancing. If a participant spends more than their total income in one round, they have a negative treasury. To balance a negative treasury a participant must return their acquired euro cents (Table 4). 3.1.4 Treatments

The experiment consists of two treatments (two income streams in Table 1). A saving treatment, starts the round with a high level of income (four or six ECU) and ends with a low level of income (zero or two ECU). The borrowing treatment has an increasing income stream that is the exact opposite of the saving treatment. It starts with a low income and ends with a high income. After one repetition of a treatment, subjects switch from borrowing to saving or vice versa.

ECU spent Euro cents bought

1 10 2 19 3 27 4 34 5 40 6 45 7 49 8 52 9 54 10 or more 55 Treasury after decision 5

Euro cents lost at the end of the round 0 or more 0 -1 ECU 12 -2 ECU 23 -3 ECU 35 -4 ECU 50 -5 ECU 72 -6 ECU 109 -7 ECU 177 -8 ECU or less 308

(17)

13 3.1.5 Optimal consumption behavior

For participants to maximize their payoff in euro cents they must smooth their consumption equally over the decisions in every round. Because of the diminishing return to ECU spent, the ratio of ECU and bought euro cents decreases (Table 3).

Assume total income in one round equals 20 ECU. If a subject smooths this income equally over the five decisions, they will spend four ECU every decision and earn 5*34= 170 euro cents. There is no better way to spend an income of 20 ECU. Table 5 shows how different spending patterns, ranging from no smoothing or overspending to good or perfect smoothing, affect earnings in a round.

The optimal strategy for both treatments is the same. If the current income in the first decision is tails, then consumption is equal in both treatments. To play optimally, subjects must take into account both current income and the rational expectation of future income. Current income is known from the coin toss. Future income can be calculated by summing the expected incomes, with equal probability. Adding current income and expected income and dividing by the number of decisions left leads to smooth consumption and maximizes the payout to the subject.

Regarding debt aversion in the experiment, specific behavior must be found, specifically, a larger deviation from the optimum in the borrowing treatment (borrow too little) than in the saving treatment (save too little).

Table 5: Spending patterns Total income is 20

ECU

Spending pattern per decision

Earnings per decision Total earnings in eurocents

No smoothing 0,0,0,0,20 0,0,0,0,55 55

Overspending total income

5,5,5,5,5 40,40,40,40,40 –72

(Treasury after decision five -5 ECU, therefore 72 is deducted)

128

Bad smoothing 0,5,5,5,5 0,40,40,40,40 160

Okay Smoothing 2,3,4,5.6 19,27,34,40,45 165

Good Smoothing 3,3,4,5,5 27,27,34,40,40 168

(18)

14

3.2 Experimental procedures

To ensure subjects have the same experimental setting, a logbook was followed (Appendix 2.). First, subjects read the instructions. In the instructions, they were introduced to income patterns, consumption decisions, payoffs and the treasury balance after the fifth decision. Subjects then had to complete a short quiz in the format of the experiment. The quiz tested the acquired

knowledge from the instructions. Subjects who finished the quiz with no mistakes then began the experiment. Subjects who failed the quiz were asked to reread the instructions and repeat the quiz until they completed it correctly.

After the subjects finished the quiz, they started round one of the experiment (either the borrowing or saving treatment). To start the round, the experimenter tossed a coin to determine the current income for that decision. Participants then checked the result of the coin toss and the income was written on the subject’s sheet. Participants had approximately 30 seconds to decide but extra time was granted for those who required it.

Once a decision was made, the researcher noted and updated the decision sheet of the participants accordingly (treasury change, treasury, euro cents bought and total euro cents). Then the next decision began. When all five decisions were made and noted, the researcher checked the treasury. If subjects spent more than their income and therefore had a negative treasury balance, euro cents were deducted accordingly and the total earnings for that round was updated. Next, round two began, which was organized similarly to round one. Only chance, the coin toss, influenced the income of participants’ every decision.

Participants received new instructions after round two. These instructions revealed a new income pattern, see Table 1 (from the saving to the borrowing treatment or vice versa). To allow for comparison between the two treatments fifteen of thirty subjects faced the saving treatment (left-hand side of Table 1) then the borrowing treatment (right-hand side). The other half of the subjects had the treatments reversed. Treatments were switched to avoid a spiked learning effect for the saving or the borrowing treatment.

(19)

15

3.3 Experimental analysis

To analyze the data from the experiment, two optimality conditions were used. These conditions follow a type of consumption smoothing: objective and subjective consumption smoothing.

The objective smoothing is determined by current income, expected future income and the treasury after the previous decision. This condition does not account for prior mistakes made by the participant and is calculated as follows for decision t (t=1,2,3,4,5):

𝑂𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑜𝑝𝑡𝑖𝑚𝑢𝑚 𝑖𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑡 (𝑡 = 1, 2, 3, 4, 5):

𝐼𝑛𝑐𝑜𝑚𝑒𝑡+ ∑ 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝐼𝑛𝑐𝑜𝑚𝑒 𝑜𝑓 𝑓𝑢𝑡𝑢𝑟𝑒 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑠 − 𝑜𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝑜𝑝𝑡𝑖𝑚𝑎𝑙 𝑇𝑟𝑒𝑎𝑠𝑢𝑟𝑦𝑡−1

𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑠 𝑙𝑒𝑓𝑡 (𝑖𝑛𝑐𝑙𝑢𝑑𝑖𝑛𝑔 𝑡)

The second condition, subjective consumption smoothing, is subject to the history of decisions, thereby smoothing current income (decision t) and future income perfectly with prior mistakes (subjective treasury). These mistakes are incorporated in the calculation after the first decision because a participant does not have a history in the first decision. If participants spend too much in an early round this is reflected in the subjective treasury and the participants should spend relatively less in the next decision. The subjective condition is calculated as follows:

𝑆𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 𝑠𝑚𝑜𝑜𝑡ℎ𝑖𝑛𝑔 𝑜𝑝𝑡𝑖𝑚𝑢𝑚 𝑖𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛 𝑡: (𝑡 = 1): 𝐼𝑛𝑐𝑜𝑚𝑒₁+ ∑ 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝐼𝑛𝑐𝑜𝑚𝑒 𝑜𝑓 𝑓𝑢𝑡𝑢𝑟𝑒 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑠 5 (𝑡 = 2, 3, 4, 5): 𝐼𝑛𝑐𝑜𝑚𝑒𝑡+ ∑ 𝐸𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝐼𝑛𝑐𝑜𝑚𝑒 𝑜𝑓 𝑓𝑢𝑡𝑢𝑟𝑒 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑠 − 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝑇𝑟𝑒𝑎𝑠𝑢𝑟𝑦𝑡−1 𝑁𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛𝑠 𝑙𝑒𝑓𝑡 (𝑖𝑛𝑐𝑙𝑢𝑑𝑖𝑛𝑔 𝑡)

In the experimental analysis, the differences between the participants’ actual spending of ECU and both optimality conditions are calculated. The median of these differences gives vital

(20)

16

insight into the behavior of the participant. If a subject perfectly smooths consumption, a median near zero is found. When the median of the differences is positive, actual ECU spent minus optimal condition, then subjects have overspent. If the median is negative, then subject have underspent.

Table 6 shows an example with an increasing income stream. This example displays the analysis of the median of the differences. In the second and third row, two types of spending are shown: overspending and underspending.

Row four shows the objective optimality condition. Rows five and six show the

subjective condition which is different for each consumption pattern. Sometimes it is optimal to spend a fractional amount, but participants can only spend integers. If a fraction is optimal then a rational participant optimizes by choosing the closest integer. When a fraction is exactly a half (1.5 for example), a rational participant is indifferent between both closest integers (1 or 2).

Rows seven and eight (nine and ten) show the differences between optimality conditions and actual ECU spent when overspending (underspending). The medians of rows seven through ten are computed. These medians give important insight into the behavior of subjects.

Significantly, positive (negative) medians indicate overspending (underspending). Larger medians, in absolute terms, imply a larger deviation from an optimality condition.

Table 6: Example of spending behavior and median analysis (continues on next page) 1. Increasing income stream

“Borrowing treatment” 2 (Tails) 1 (Heads) 4 (Tails) 3 (Heads) 6 (Tails)

2. Overspending (ECU spent) 4 5 4 2 1

3. Underspending (ECU spent) 1 2 2 4 7

4. Objective 3.2 2.95 3.28 2.78 3.78

5. Subjective 1 “overspend” 3.2 2.75 2.33 1 1

(21)

17

7. Overspend minus objective 0.8 2.05 0.72 (Median)

-.78 -2.78

8. Overspend minus subjective 1 .8 2.25 1.67 1 (Median)

0

9. Underspend minus objective -1.2 (Median)

-1.95 -1.28 1.22 3.22

10. Underspend minus subjective 2 -1.2 (Median)

-2.25 -2.33 -1 0

Table 6: continued 3.3.1 Overspending

Subjects are expected to overspend. To investigate this issue a sign test is used and is conducted on the subjective and objective median differences (Appendix 3.). To measure the deviation of the increasing (decreasing) income stream, the data of subjects who played the borrowing (saving) treatment in the second or fourth round are combined.

Table 7 shows a difference between the optimality conditions. Considering the objective optimality condition, there is a significant deviation in the first round of borrowing because of underspending (medians tend to be negative). However, this result is not found in the second round of borrowing.

Furthermore, there is no indication that there is overspending in the saving treatment. Regarding the subjective optimality condition, both borrowing rounds indicate a significant deviation from the optimal behavior due to underspending. Additionally, in the saving rounds, there is a significant deviation from zero, because of overspending (medians tend to be positive). The overspending hypothesis does not hold for the objective optimality condition. For the subjective condition the hypothesis only holds for the saving treatment but not for the borrowing treatment. This result indicates possible debt averse behavior, as defined previously. Because of these results, this thesis will focus on the subjective optimality condition.

(22)

18

Table 7: Probabilities of sign tests * p<0.05; H0 is rejected 3.3.2 Learning

To measure if subjects learn with experience, median differences of the subjective condition in each round are summed up (graph 1). Table 6 shows an example of these medians for one participant in one round. Graph 1 shows the aggregate medians (in absolute terms) in every round for all participants.

The aggregate medians are the highest in the first round and lower in subsequent rounds. A linear line is estimated with a 95% confidence interval. If medians approach zero, subjects deviate less. Although subjects improve on average, there is no significant downward trend that indicates learning. This is displayed in the graph because the confidence intervals continually overlap. Subjects do not learn with experience, which is not in line with previous research.

Sign test hypothesis Objective optimality

condition 1st_round Borrowing 2nd_round Borrowing 1st_round Saving 2nd_round Saving H0: Median = 0, H1: Median < 0 0.00* 0.29 N/A N/A H0: Median = 0,

H1: Median > 0 N/A N/A 0.29 0.29

Subjective optimality condition H0: Median = 0, H1: Median < 0 0.00* 0.00* N/A N/A H0: Median = 0, H1: Median > 0 N/A N/A 0.00* 0.02*

(23)

19

Graph 1: Learning across rounds 3.3.3 Debt aversion

As discussed earlier, there is no significant learning and results of the objective condition are disregarded. Therefore, there is a focus on the subjective condition and comparison between each round is possible. To be able to compare the average median deviations, confidence intervals are constructed. Normality of medians was checked before the construction of these intervals. Two rounds are disregarded in this analysis, the second borrowing round and the first saving round (Appendix 4.). See Appendix 5 for a nonparametric bootstrap estimation. The result of this estimation is similar to the following analysis.

The graph below gives key insight into the debt aversion of subjects. The graph shows the confidence interval around the average median difference given the treatment. It is clear that the confidence intervals do not overlap with 0, visually supporting the conclusions of section 3.3.1. Using the average of the median differences in the graph below, it is concluded that the deviation from the subjective condition is not greater in the borrowing treatment than in the

0 1 0 2 0 3 0 4 0 5 0 S u m o f a b s o lu te m e d ia n s 1 2 3 4 Rounds 95% CI Fitted values

(24)

20

saving treatment. Therefore, Dutch students are not debt averse and the third hypothesis does not hold. These results are in contrast to previous literature (Meissner, 2016).

Graph 2: 95% confidence intervals of medians

4. Regression results

To extend upon previous research, I investigate whether a debt averse individual is more likely to avoid debt in real life. This section will test whether the results from the experiment are

externally valid. To allow for this validation, variables need to be defined. Firstly, debt aversion is described. Secondly, two variables of actual debt behavior are discussed. Lastly, regression results are displayed. A summary of all variables is included in Appendix 5.

4.1 Debt aversion in the experiment

There are requirements to identify individual subjects as debt averse. These conditions mimic the definition of debt aversion. Using the subjective optimality condition a subject is debt averse if they meet the following requirements:

-1 .5 -1 -. 5 0 .5 1 A v e ra g e m e d ia n d e v ia ti o n

Subjective Borrowing 1st round Subjective Saving 2nd round 95% confidence intervals

(25)

21

1. A subject does not underspend in the saving treatment (median ≥ 0) and does not overspend in the borrowing treatment (median ≤ 0)

2. A subject deviates more, in absolute terms, in the borrowing treatment than in the saving treatment (borrowing treatment has a larger median in absolute value)

Participants must not overspend in the borrowing treatment. If a participant overspends in the borrowing treatment, for example by taking on too much debt, then it is safe to assume that this participant is not debt averse. This definition of debt aversion identifies participants who have anomalous behavior in both treatments but deviate more from the subjective condition in the borrowing treatment than in the saving treatment.

Using these requirements, eleven of thirty subjects are debt averse in the experiment. To incorporate these subjects in a regression analysis a dummy variable is created.

4.2 Real debt behavior

This section will focus on the potential relation between debt aversion from the experiment and real debt behavior. In the survey, two potential proxies for real debt behavior were included. “Monthly debt increase" in euros is the first variable that is tested using a robust OLS-regression. Then “fraction debt of income" each month

is assessed by making use of a robust probit regression.

Table 8 indicates correlations with the other control variables of the survey. The Table suggests that “monthly debt increase" and “fraction debt of income" are similar variables, due to a high correlation of 0.93. Furthermore, variables are indicated (with *) that could have predictive value for the variables for actual debt behavior.

To interpret Table 8, two variables are discussed: “Debt aversion" and

“University". Debt aversion has a negative Table 8: Correlation table Monthly debt increase Fraction Debt of income Monthly debt increase 1 Fraction Debt of income 0.93 1 Debt aversion -0.27* -0.27* Male -0.19* -0.05 Job -0.05 -0.10 Support -0.19* -0.22* High Earner 0.03 0.12* University 0.19* 0.18*

(26)

22

sign. This indicates that subjects who were identified as debt averse have a lower monthly debt increase and a smaller fraction debt of income than non-debt averse subjects. University has an opposite effect because of the positive correlation between these variables. University students tend to have an larger monthly debt increase and a higher fraction of debt.

4.3 Results

First, this section focuses on “monthly debt increase" as the independent variable in a robust OLS-regression. Robust regression is used to allow for heteroscedasticity in variables.

Previously, several control variables were identified. The models below include variables that have a high correlation with “monthly debt increase". Two regression models are displayed:

(1) Monthly debt increase = β0 + β1

* (Debt aversion) + Ɛ0

(2) Monthly debt increase = β0 + β1

* (Debt aversion) + β2 *

(University) + β3 * (Job) + β4 *

(Support) + Ɛ0

Table 9 shows the results of the regression models. Regarding the first model, there is no significant relation between debt aversion and monthly debt increase. This model is extended with three variables: University, Male and Support. In Model (2), there is again no

significant relation (P-value < 0.05) between any of the independent variables and the dependent variable.

If a P-value of 0.10 is used, debt aversion Table 9: OLS-regression results

Model (1) (2) Monthly debt increase Monthly debt increase Debt aversion -155.5 -170.4* (-1.65) (-1.91) University 168.7 -1.54 Male -124.5 (-1.18) Support -116.5 (-1.19) Constant 355.5** 385.8** -4.83 -2.77 Observations R2 30 0.0707 30 0.2066 t statistics in parentheses * p<0.10, ** p<0.05

(27)

23

impacts monthly debt increase. If subjects are debt averse, they are estimated to accumulate 170,40 euro less debt.

By comparing the R2(coefficient of determination), the proportion of variance in the dependent variable (Monthly debt increase) is predicted by the independent variables (Debt aversion, University, Male and Support). Model 1 (R2 of 0.0707) performs worse than model 2 (R2 of 0.2066). Model 2 provides a better explanation of the variation in the monthly debt increase.

Secondly, this section investigates “fraction of debt" as an independent variable. Four probit models are used to allow for fractional regression.

(1)Pr (Fraction of debt = 1 | Debt aversion) = Φ (β0 + β1 * (Debt aversion))

(2)Pr (Fraction of debt = 1 | Debt aversion, University, High Earner) = Φ (β0 + β1 * (Debt

aversion) + β2 * (University) + β3 * (High Earner))+

(3) Pr (Fraction of debt = 1 | Debt aversion, University, Support, High Earner) = Φ (β0 + β1 *

(Debt aversion) + β2 * (University) + β3 * (Support))

(4) Pr (Fraction of debt = 1 | Debt aversion, University, Support, High Earner) = Φ (β0 + β1 *

(Debt aversion) + β2 * (University) + β3 * (Support) + β4 * (High Earner))

The result of the regression models is displayed in Table 10, which shows a similar result as before. Importantly, model (2) shows a significant relation (P-value < 0.05) between debt

aversion and the fraction of debt. Therefore it is concluded that the findings in the experiment are externally valid. If subjects are debt averse, they are expected to have a lower fraction of debt by 16.7% (Margins Table in Appendix 6.). If other variables are added to the regression ((3) and (4)), debt aversion loses its significance.

The models are evaluated by comparing the pseudo-R2s. Similar to the previous R2, the pseudo R2 indicates how well the dependent variable (fraction of debt) is predicted by the

independent variables (Debt aversion, University, Support and High Earner). By adding variables to model 1, the pseudo R2_{continues to improve. While model 2 has a significant independent}

(28)

24

variable (debt aversion), model 4 has the highest pseudo R2. Therefore, the best prediction of the fraction of debt is model 4.

Model (1) (2) (3) (4) Fraction of Debt Fraction of Debt Fraction of Debt Fraction of Debt Debt aversion -0.452 -0.523** -0.537* -0.536* (-1.64) (-1.97) (-1.94) (-1.93) University 0.404 0.456 0.506 (-1.31) (-1.25) (-1.58) Support -0.37 -0.385 (-1.31) (-1.30) High Earner 0.0384 -0.0814 (-0.14) (-0.27) Constant -0.446** -0.755** -0.536 -0.507 (-2.52) (-1.99) (-1.36) (-1.13) Observations Pseudo R2 30 0.0212 30 0.0362 30 0.0509 30 0.0513 t statistics in parentheses * p<0.10, ** p<0.05

Table 10: Probit-regression results

5. Discussion

5.1 Main findings

Overall, the experiment does not find debt aversion. Deviation from the subjective optimality condition in the saving and borrowing treatment is found. It is concluded that subjects deviate from the subjective optimal consumption path. In the saving treatment, subjects tend to overspend and underspend in the borrowing treatment.

(29)

25

When subjects are analyzed individually (subjects are identified as debt averse based on their behavior in the experiment) a dependent variable can be included in the regression models, namely debt aversion. It is found that debt averse subjects in the experiment tend to be debt averse in real life. Weak evidence (P-value<0.1) suggests that they accumulate less debt and strong evidence (P-value<.05) indicates that subjects are less dependent on debt as a part of their income.

5.2 Implications

Previous literature (Carbone, & Duffy, 2014; Meissner, 2016; Ballinger, Palumbo, & Wilcox, 2003) found that subjects learn with experience. Section 3.3.2 describes this learning effect. Although subjects improve on average across rounds there is no significant improvement. This is probably due to the quiz that is required before beginning the experiment. This implies that the goal of the quiz is achieved: subjects understand the instructions and the experiment.

Hypotheses were defined from literature in the introduction. Overspending is expected (Brown, Chua, & Camerer, 2009; Thaler, 1980). In the experimental analysis (section 3.3.1) subjects, tend to overspend in the saving treatment. When subjects are confronted with the borrowing treatment, they typically underspend. These results are supported by Meissner (2016).

Meissner (2016) first described experimental debt aversion. To test the internal validity of this research a simplified experiment is used. This experiment included similar treatments. Meissner (2016) found debt aversion based on a comparison between a subject's experimental behavior and optimality conditions. This thesis used the same conditions to compare the results. In section 3.3.3 it is concluded that there is no evidence for debt aversion in the experiment.

Furthermore, chapter four uses a novel approach to test the external validity of the

experiment by considering proxies for real-life debt behavior. Debt averse subjects are identified based on their behavior. This definition of debt aversion is new in literature. As a result, eleven of thirty were identified as debt averse. Several regression models were tested. These models revealed varied results. In the OLS-regression weak significance (P-value < 0.1) is found with respect to debt aversion to monthly increasing debt. The stronger significance is found in the fractional probit regression. Model (2) in regression Table 2 revealed a significant relation (P-value < 0.05) between debt aversion and debt as a fraction of income. It is estimated that subjects who are debt averse have a lower fraction of debt (16,7% less) than non-debt averse subjects.

(30)

26

Consequently, despite being unable to replicate the results of Meissner (2016), the

experiment and the accompanying regression models show that debt aversion predicts actual debt behavior. This extends current literature. This implies that the demand for debt is partially

predicted by the level of debt aversion in the population, which suggests that the Dutch

government (or educational institutions) can use this test to determine whether students are debt averse and to keep track of the changes in debt aversion.

When individuals are enrolled as full-time students, they have less time to work and are therefore more dependent on debt, to support their lifestyle (Werf, Schonewille, & Stoo, 2017). Importantly, if individuals become more debt averse, they are more likely to avoid debt and full-time education. This can cause shortages in the number of highly educated people and amount of future human capital. Currently, there is a shortage of university-trained people, such as

technicians (ICT) and doctors (“Personeelstekort”, 2017). Moreover, people who choose work over higher education contribute less and pay less tax because of a lower-income in the future. Therefore, society can improve if the benefits of debt increase.

The role of governments in the debt accumulation of students is crucial. They can set up awareness campaigns so that students can recognize their anomalous behavior. For example, governments could explain that borrowing or investing in the early stages of life is important for personal development and future prospects. Governments could intervene in other ways, such as making debt more attractive by lowering interest rates or improving fiscal benefits once

individuals are working. 5.3 Limitations

The results of the experiment do not indicate debt aversion for participants overall. Therefore, it is concluded that Dutch students are not debt averse. But from an individual perspective, debt aversion significantly affects the amount of monthly accumulated debt and debt as a fraction of income. Debt averse students use less debt. The possible problems and limitations are discussed below.

The first possible problem that arises, is the sample. Thirty Dutch students participated in the experiment. If more students participated in the experiment, the size of the standard deviation of average median difference could be smaller. This would lead to more accurate conclusions about debt aversion, or potentially saving aversion. Furthermore, increasing the sample size would lead to better predictions in the regression models. This increase could give different

(31)

27

results as the population of Dutch students would be more accurately represented. Moreover, if the number of observations is increased then more independent variables can be included in the regression without it losing its power.

Furthermore, students should be less debt averse in comparison to individuals who did not go to university. Therefore, the sample of students can give a contrasting result in

comparison to non-students. Moreover, it is expected that debt aversion changes over a lifetime. As people get older the expected benefits of the currently accumulated debt decrease as the time to repay debt diminishes. Aging can affect the willingness to educate and ability to learn

negatively, increasing debt averse behavior. Consequently, the results of this thesis are limited to Dutch students.

The second point of issue is the experiment itself. Subjects earned 248 euro cents on average. This incentive is possibly too weak and therefore does not reflect real behavior.

Furthermore, the experiment might be oversimplified. Participants can only spend integers, while fractions are optimal. The experiment may steer behavior. The strict consequences of having debt after the fifth round might have caused the participants to be conservative in their debt accumulation and consequently their consumption smoothing. Additionally, if subjects use a rule of thumb such as: “consumption is equal to current income in each decision” then they will not use their treasury causing them to underspend (overspend) in the borrowing (saving) treatment. Moreover, the experiment might not reflect reality. Individuals have one life and it is not possible to learn across lifetimes. Since the experiment has multiple rounds, subjects can learn from past mistakes and improve their earnings. Additionally, this experiment is conducted with pen-and-paper, which reduces the number of decisions in the experiment. If the experiment is done on a computer, rounds could have been extended and more observations made about individual subjects.

A third caveat is related to the regression results, specifically the definition of debt aversion. This novel definition may be too relaxed. Subjects are defined as debt averse if they deviate more from optimal behavior in the borrowing treatment than in the saving treatment. Results might differ if a stricter definition of debt aversion is used, for example, that the deviation must be twice as large.

Another possible problem stems from the survey questions. Despite the anonymity in the experiment, debt has a negative association in Dutch society so subjects might feel shame for

(32)

28

their debt behavior (Stadhouders, 2018). Therefore, subjects may indicate a lower monthly debt increase and a lower debt as a fraction of income. Van der Werf, Schonewille, & Stoo (2018), found that Dutch students accumulate 528 euro each month on average, whereas this study finds an average of 298.50 euro.

5.4 Future research

For future research, it may be useful to adapt the experiment and allow for rounds with more decisions, preferably, one round for the borrowing treatment and one round for the saving treatment. To avoid a learning effect follow-up research should include a quiz. By increasing the number of decisions in one round more specific behavior can be found. In the saving (borrowing) treatment, subjects would need to save (borrow) in more decisions at the start of the round to be able to maintain consumption in later decisions. This can lead to a greater deviation from the optimality conditions and may show more debt averse behavior in the experiment.

Another topic for future research is adjusting the experiment to different types of individuals. For example, comparing different age-groups or people in different stages of their life, such as full-time workers and pensioners. These groups might differ in their debt averse behavior.

Other future research topics might include the impact of debt averse individuals on society, the extent to which it is possible to quantify the effects of debt aversion, or, how large the loss of potential human capital is if individuals do not want to invest in education.

Additionally, is it possible to adjust the current debt system or increase awareness to reduce the negative effects of debt aversion. Moreover, it would be interesting to research why individuals are debt averse. Is debt aversion related to risk aversion or other characteristics such as ability, origin, ethnicity, age, gender, wealth or income class.

6. Conclusion

The goal of this thesis was to determine to what extent debt aversion affects real debt behavior among Dutch students. To answer this question an experiment, based on Meissner’s experiment (2016), was conducted. This experiment has a saving and a borrowing treatment. To maximize payoff, subjects need to smooth consumption over multiple decisions. In the saving (borrowing) treatment subjects must save (borrow) at the start of the round to play optimally. Both treatments

(33)

29

indicate a significant deviation from the subjective optimal path. In the saving (borrowing) treatment, subjects have overspent (underspent). But there is no indication that subjects deviate more in the borrowing treatment than in the saving treatment, therefore it is concluded that, overall, subjects are not debt aversive.

Despite these results, the experimental findings were used to identify individual subjects as debt averse. Two independent variables were identified and surveyed in the last part of the experiment, namely, “monthly debt increase” and “debt as a fraction of income”. These variables were included in multiple regression models with varied results. Answering the research

question, model (2) of the probit regression table showed a significant result, specifically, when subjects are identified as debt averse, they are expected to have less debt as a part of their income (16.7% less).

This approach was new in literature therefore, results should be carefully interpreted. Future research should repeat the experiment to verify the results of this thesis. Further research on this topic should use larger samples to make conclusions more accurate.

This thesis contributes to a better understanding of debt aversion and the impact on real debt behavior. However, more research is needed to verify results, analyze the impact of debt aversion on society and determine why individuals are debt averse. The experiment can be improved by avoiding the steering of behavior and increasing the number of decisions in one round.

(34)

30

References

Ando, A., & Modigliani, F. (1963). The "Life Cycle" Hypothesis of Saving: Aggregate Implications and Tests. The American Economic Review, 53(1), 55-84.

Ballinger, T. P., Palumbo, M. G., & Wilcox, N. T. (2003). Precautionary saving and social learning across generations: an experiment. The Economic Journal 113(490).

Bernheim, B. D. (1987). Dissaving after Retirement: Testing the Pure Life Cycle Hypothesis

Pension Economics University of Chicago Press, 237-280.

https://www.nber.org/chapters/c6861.pdf

Brown, A. L., Chua, Z. E., & Camerer, C. F. (2009). Learning and Visceral Temptation in Dynamic Saving Experiments. The Quarterly Journal of Economics, 124(1), 197–231. Browning, M., & Lusardi, A. (1996). Household Saving: Micro Theories and Micro Facts.

Journal of Economic Literature, 34(4), 1797-1855.

Caetano, G., Palacios, M., & Patrinos, H. (2011). Measuring aversion to debt: An experiment among student loan candidates. World Bank Policy Research Working Paper Series. Carbone, E., & Duffy, J. (2014). Lifecycle consumption plans, social learning and external

habits: experimental evidence. Journal of Economic Behavior & Organization, 106, 413-427.

Cunningham, A. F., & Santiago, D. A. (2008). Student Aversion to Borrowing: Who Borrows and Who Doesn't. Washington: Institute for Higher Education Policy and Excellence in

Education.

Eckel, C.C., Johnson, C., Montmarquette, C., & Rojas, C. (2016). Debt Aversion and the

Demand for Loans for Postsecondary Education. Public Finance Review 35(2), 233-262. Friedman, M. (1957). The permanent income hypothesis. American Economic Review, 48, 80-91. Hall, R. & Mishkin, F. (1982). The sensitivity of consumption to transitory income: estimates

from panel data on households. Econometrica, 50(2), 461–81.

Heckman, J. (1974). Life Cycle Consumption and Labor Supply: An Explanation of the Relationship between Income and Consumption Over the Life Cycle. The American Economic Review, 64(1), 188-194.

Jiménez-Buedo, M., & Miller, L. M. (2010). Why a Trade-Off? The Relationship between the External and Internal Validity of Experiments. Theoria, 25(3), 301-321.

(35)

31

Keynes, J. (1937). The General Theory of Employment. The Quarterly Journal of

Economics, 51(2), 209-223.

Kimball, M. (1990). Precautionary saving in the small and in the large. Econometrica, 58(1), 53– 73.

Levitt, S. D., & List, A. J. (2007). What Do Laboratory Experiments Measuring Social Preferences Reveal About the Real World? Journal of Economic Perspectives, 21(2), 153-174.

Mankiw, N. G. (2012). Macroeconomics: Eighth Edition. W.H.Freeman & Co Ltd Meissner, M. (2016). Intertemporal consumption and debt aversion: an experimental.

Experimental Economics, 19(2), 281-298.

Morduch, J. (1995). Income Smoothing and Consumption Smoothing. Journal of Economic

Perspectives, 9(3), 103-114.

Parker, J. A., Brunnermeier, M. K., & Papakonstantinou, F. (2016). Optimal Time-Inconsistent Beliefs: Misplanning, Procrastination, and Commitment. Management Science 63(5). Prelec, D., & Loewenstein, G. (1998). The Red and the Black: Mental Accounting of Savings

and Debt. Marketing Science, 17(1), 4-89.

Schram, A. (2005). Artificiality: The tension between internal and external validity in economic experiments. Journal of Economic Methodology, 12(2), 225-237.

Shapiro, M., & Slemrod, J. (2003). Consumer Response to Tax Rebates. The American

Economic Review, 93(1), 381-396.

Smith, A. (1776). The wealth of nations.

Souleles, N. (1999). The Response of Household Consumption to Income Tax Refunds. The

American Economic Review, 89(4), 947-958.

Sulkunen, P. (1978). Individual Consumption in Capitalism. An Exercise in the Logic of Capital. Acta Sociologica, 21(1), 35-46.

Thaler, R. (1980) Toward a positive value of consumer choice. Journal of Economic Behavior &

Organization, 1 (1), 39–60.

Van der Werf, M., Schonewille, G., & Stoo, R. (2017). Studentenonderzoek 2017. Nibud. Retrieved from https://www.nibud.nl/wp-content/uploads/Nibud-Studentenonderzoek-2017.pdf

(36)

32

Walters, D., Erner, C., Fox, C., Scholten, M., Read, D., & Trepel C. (2016). Debt Aversion: Anomalous in Theory, Advantageous in Practice. Advances in Consumer Research 44, 179-184.

Websites

“Beloftes” (2018). Het kabinet moet eerst eens werk maken van die beloftes bij de invoering van het leenstelsel. Retrieved from: https://www.trouw.nl/opinie/het-kabinet-moet-eerst-eens-werk-maken-van-die-beloftes-bij-de-invoering-van-het-leenstelsel~a2a523bf/

“Minder studenten” (2016).Minder studenten na invoering leenstelsel. Retrieved from https://www.trouw.nl/home/minder-studenten-na-invoering-leenstelsel~a6d1186e/ “Personeelstekort” (2017). In 2022 groot personeelstekort in techniek, zorg en onderwijs.

NOS.nl. Retrieved from

https://nos.nl/artikel/2205872-in-2022-groot-personeelstekort-in-techniek-zorg-en-onderwijs.html

Stadhouders, S. (2018). Studenten steken zich gierend in de schulden. Followthemoney.nl Retrieved from https://www.ftm.nl/artikelen/kasboekje-studentenleningen?share=1

(37)

33

Appendix

1. Instructions of the experiment

Introduction

The experiment in which you are participating is part of a thesis project. Based on your decisions and chance you can earn money. The rules and instructions are the same for every participant.

The duration of the experiment is around 45 minutes.

You are not allowed to talk or exchange information with other participants during the

experiment. Please raise your hand if you have a question. The experimenter will then come to your desk and will answer your question. Do not ask your questions out loud.

Please do not write your name anywhere so your decisions are anonymous. Please read the instructions carefully.

Overview

First you will have time to read the instructions. Then you will answer a quiz to make sure you understand the instructions. If you finished the quiz please raise your hand, the experimenter will check your answers.

When you completed the quiz, the experiment will start. The experiment has four rounds. Each round consists of five decisions. Before every decision, a coin toss (Heads or

tails) determines your income. You receive income in Experimental Currency Units (ECU). Your income is announced out loud and written on your decision sheet.

You decide how many ECU you spend to buy eurocents. You accumulate eurocents across decisions and rounds. At the end of the experiment you receive the eurocents gathered across all rounds divided by two. After every participants' decision is noted by the experimenter the next decision starts.

After the experiment you will answer a short survey.

If at any time during the experiment you need further explanation, please raise your hand.

Instruction round 1

This experiment requires you to make decisions on how to spend your income. During a round you make five decisions. These decisions determine your payoff.